Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/svcsock.c | |
3 | * | |
4 | * These are the RPC server socket internals. | |
5 | * | |
6 | * The server scheduling algorithm does not always distribute the load | |
7 | * evenly when servicing a single client. May need to modify the | |
f6150c3c | 8 | * svc_xprt_enqueue procedure... |
1da177e4 LT |
9 | * |
10 | * TCP support is largely untested and may be a little slow. The problem | |
11 | * is that we currently do two separate recvfrom's, one for the 4-byte | |
12 | * record length, and the second for the actual record. This could possibly | |
13 | * be improved by always reading a minimum size of around 100 bytes and | |
14 | * tucking any superfluous bytes away in a temporary store. Still, that | |
15 | * leaves write requests out in the rain. An alternative may be to peek at | |
16 | * the first skb in the queue, and if it matches the next TCP sequence | |
17 | * number, to extract the record marker. Yuck. | |
18 | * | |
19 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
20 | */ | |
21 | ||
172589cc | 22 | #include <linux/kernel.h> |
1da177e4 LT |
23 | #include <linux/sched.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/fcntl.h> | |
26 | #include <linux/net.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/inet.h> | |
29 | #include <linux/udp.h> | |
91483c4b | 30 | #include <linux/tcp.h> |
1da177e4 LT |
31 | #include <linux/unistd.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/netdevice.h> | |
34 | #include <linux/skbuff.h> | |
b41b66d6 | 35 | #include <linux/file.h> |
7dfb7103 | 36 | #include <linux/freezer.h> |
1da177e4 LT |
37 | #include <net/sock.h> |
38 | #include <net/checksum.h> | |
39 | #include <net/ip.h> | |
b92503b2 | 40 | #include <net/ipv6.h> |
b7872fe8 | 41 | #include <net/tcp.h> |
c752f073 | 42 | #include <net/tcp_states.h> |
1da177e4 LT |
43 | #include <asm/uaccess.h> |
44 | #include <asm/ioctls.h> | |
45 | ||
46 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 47 | #include <linux/sunrpc/clnt.h> |
1da177e4 | 48 | #include <linux/sunrpc/xdr.h> |
c0401ea0 | 49 | #include <linux/sunrpc/msg_prot.h> |
1da177e4 LT |
50 | #include <linux/sunrpc/svcsock.h> |
51 | #include <linux/sunrpc/stats.h> | |
52 | ||
360d8738 | 53 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
54 | |
55 | ||
56 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 57 | int *errp, int flags); |
1da177e4 LT |
58 | static void svc_udp_data_ready(struct sock *, int); |
59 | static int svc_udp_recvfrom(struct svc_rqst *); | |
60 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 61 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 62 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 63 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 64 | |
b700cbb1 TT |
65 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
66 | struct sockaddr *, int, int); | |
ed07536e PZ |
67 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
68 | static struct lock_class_key svc_key[2]; | |
69 | static struct lock_class_key svc_slock_key[2]; | |
70 | ||
0f0257ea | 71 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
72 | { |
73 | struct sock *sk = sock->sk; | |
02b3d346 | 74 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
75 | switch (sk->sk_family) { |
76 | case AF_INET: | |
77 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
78 | &svc_slock_key[0], |
79 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
80 | &svc_key[0]); | |
ed07536e PZ |
81 | break; |
82 | ||
83 | case AF_INET6: | |
84 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
85 | &svc_slock_key[1], |
86 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
87 | &svc_key[1]); | |
ed07536e PZ |
88 | break; |
89 | ||
90 | default: | |
91 | BUG(); | |
92 | } | |
93 | } | |
94 | #else | |
0f0257ea | 95 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
96 | { |
97 | } | |
98 | #endif | |
99 | ||
1da177e4 LT |
100 | /* |
101 | * Release an skbuff after use | |
102 | */ | |
5148bf4e | 103 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 104 | { |
5148bf4e | 105 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
106 | |
107 | if (skb) { | |
57b1d3ba TT |
108 | struct svc_sock *svsk = |
109 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 110 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
111 | |
112 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
57b1d3ba | 113 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 | 114 | } |
1da177e4 LT |
115 | } |
116 | ||
b92503b2 CL |
117 | union svc_pktinfo_u { |
118 | struct in_pktinfo pkti; | |
b92503b2 | 119 | struct in6_pktinfo pkti6; |
b92503b2 | 120 | }; |
bc375ea7 DM |
121 | #define SVC_PKTINFO_SPACE \ |
122 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
123 | |
124 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
125 | { | |
57b1d3ba TT |
126 | struct svc_sock *svsk = |
127 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
128 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
129 | case AF_INET: { |
130 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
131 | ||
132 | cmh->cmsg_level = SOL_IP; | |
133 | cmh->cmsg_type = IP_PKTINFO; | |
134 | pki->ipi_ifindex = 0; | |
135 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
136 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
137 | } | |
138 | break; | |
5a05ed73 | 139 | |
b92503b2 CL |
140 | case AF_INET6: { |
141 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
142 | ||
143 | cmh->cmsg_level = SOL_IPV6; | |
144 | cmh->cmsg_type = IPV6_PKTINFO; | |
145 | pki->ipi6_ifindex = 0; | |
146 | ipv6_addr_copy(&pki->ipi6_addr, | |
147 | &rqstp->rq_daddr.addr6); | |
148 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
149 | } | |
150 | break; | |
b92503b2 CL |
151 | } |
152 | return; | |
153 | } | |
154 | ||
1da177e4 LT |
155 | /* |
156 | * Generic sendto routine | |
157 | */ | |
0f0257ea | 158 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) |
1da177e4 | 159 | { |
57b1d3ba TT |
160 | struct svc_sock *svsk = |
161 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1da177e4 LT |
162 | struct socket *sock = svsk->sk_sock; |
163 | int slen; | |
bc375ea7 DM |
164 | union { |
165 | struct cmsghdr hdr; | |
166 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
167 | } buffer; | |
168 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
169 | int len = 0; |
170 | int result; | |
171 | int size; | |
172 | struct page **ppage = xdr->pages; | |
173 | size_t base = xdr->page_base; | |
174 | unsigned int pglen = xdr->page_len; | |
175 | unsigned int flags = MSG_MORE; | |
5216a8e7 | 176 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
177 | |
178 | slen = xdr->len; | |
179 | ||
180 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
181 | struct msghdr msg = { |
182 | .msg_name = &rqstp->rq_addr, | |
183 | .msg_namelen = rqstp->rq_addrlen, | |
184 | .msg_control = cmh, | |
185 | .msg_controllen = sizeof(buffer), | |
186 | .msg_flags = MSG_MORE, | |
187 | }; | |
188 | ||
189 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
190 | |
191 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
192 | goto out; | |
193 | } | |
194 | ||
195 | /* send head */ | |
196 | if (slen == xdr->head[0].iov_len) | |
197 | flags = 0; | |
44524359 N |
198 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
199 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
200 | if (len != xdr->head[0].iov_len) |
201 | goto out; | |
202 | slen -= xdr->head[0].iov_len; | |
203 | if (slen == 0) | |
204 | goto out; | |
205 | ||
206 | /* send page data */ | |
207 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
208 | while (pglen > 0) { | |
209 | if (slen == size) | |
210 | flags = 0; | |
e6242e92 | 211 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
212 | if (result > 0) |
213 | len += result; | |
214 | if (result != size) | |
215 | goto out; | |
216 | slen -= size; | |
217 | pglen -= size; | |
218 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
219 | base = 0; | |
220 | ppage++; | |
221 | } | |
222 | /* send tail */ | |
223 | if (xdr->tail[0].iov_len) { | |
44524359 N |
224 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
225 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 226 | & (PAGE_SIZE-1), |
1da177e4 LT |
227 | xdr->tail[0].iov_len, 0); |
228 | ||
229 | if (result > 0) | |
230 | len += result; | |
231 | } | |
232 | out: | |
ad06e4bd | 233 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 234 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 235 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
236 | |
237 | return len; | |
238 | } | |
239 | ||
80212d59 N |
240 | /* |
241 | * Report socket names for nfsdfs | |
242 | */ | |
e7942b9f | 243 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 244 | { |
017cb47f CL |
245 | const struct sock *sk = svsk->sk_sk; |
246 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
247 | "udp" : "tcp"; | |
80212d59 N |
248 | int len; |
249 | ||
017cb47f | 250 | switch (sk->sk_family) { |
e7942b9f CL |
251 | case PF_INET: |
252 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f CL |
253 | proto_name, |
254 | &inet_sk(sk)->rcv_saddr, | |
255 | inet_sk(sk)->num); | |
80212d59 | 256 | break; |
58de2f86 CL |
257 | case PF_INET6: |
258 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f CL |
259 | proto_name, |
260 | &inet6_sk(sk)->rcv_saddr, | |
261 | inet_sk(sk)->num); | |
80212d59 N |
262 | break; |
263 | default: | |
e7942b9f | 264 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 265 | sk->sk_family); |
80212d59 | 266 | } |
e7942b9f CL |
267 | |
268 | if (len >= remaining) { | |
269 | *buf = '\0'; | |
270 | return -ENAMETOOLONG; | |
80212d59 N |
271 | } |
272 | return len; | |
273 | } | |
274 | ||
8435d34d CL |
275 | /** |
276 | * svc_sock_names - construct a list of listener names in a string | |
277 | * @serv: pointer to RPC service | |
278 | * @buf: pointer to a buffer to fill in with socket names | |
279 | * @buflen: size of the buffer to be filled | |
280 | * @toclose: pointer to '\0'-terminated C string containing the name | |
281 | * of a listener to be closed | |
282 | * | |
283 | * Fills in @buf with a '\n'-separated list of names of listener | |
284 | * sockets. If @toclose is not NULL, the socket named by @toclose | |
285 | * is closed, and is not included in the output list. | |
286 | * | |
287 | * Returns positive length of the socket name string, or a negative | |
288 | * errno value on error. | |
289 | */ | |
290 | int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, | |
291 | const char *toclose) | |
80212d59 | 292 | { |
b41b66d6 | 293 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
294 | int len = 0; |
295 | ||
296 | if (!serv) | |
297 | return 0; | |
e7942b9f | 298 | |
aaf68cfb | 299 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 300 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
e7942b9f CL |
301 | int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); |
302 | if (onelen < 0) { | |
303 | len = onelen; | |
304 | break; | |
305 | } | |
306 | if (toclose && strcmp(toclose, buf + len) == 0) | |
b41b66d6 N |
307 | closesk = svsk; |
308 | else | |
309 | len += onelen; | |
80212d59 | 310 | } |
aaf68cfb | 311 | spin_unlock_bh(&serv->sv_lock); |
e7942b9f | 312 | |
b41b66d6 | 313 | if (closesk) |
5680c446 N |
314 | /* Should unregister with portmap, but you cannot |
315 | * unregister just one protocol... | |
316 | */ | |
7a182083 | 317 | svc_close_xprt(&closesk->sk_xprt); |
37a03472 N |
318 | else if (toclose) |
319 | return -ENOENT; | |
80212d59 N |
320 | return len; |
321 | } | |
24c3767e | 322 | EXPORT_SYMBOL_GPL(svc_sock_names); |
80212d59 | 323 | |
1da177e4 LT |
324 | /* |
325 | * Check input queue length | |
326 | */ | |
0f0257ea | 327 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 328 | { |
1da177e4 LT |
329 | struct socket *sock = svsk->sk_sock; |
330 | int avail, err; | |
331 | ||
e6242e92 | 332 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
333 | |
334 | return (err >= 0)? avail : err; | |
335 | } | |
336 | ||
337 | /* | |
338 | * Generic recvfrom routine. | |
339 | */ | |
0f0257ea TT |
340 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
341 | int buflen) | |
1da177e4 | 342 | { |
57b1d3ba TT |
343 | struct svc_sock *svsk = |
344 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
345 | struct msghdr msg = { |
346 | .msg_flags = MSG_DONTWAIT, | |
347 | }; | |
348 | int len; | |
1da177e4 | 349 | |
260c1d12 TT |
350 | rqstp->rq_xprt_hlen = 0; |
351 | ||
1ba95105 CL |
352 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
353 | msg.msg_flags); | |
1da177e4 | 354 | |
1da177e4 | 355 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 356 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
357 | return len; |
358 | } | |
359 | ||
360 | /* | |
361 | * Set socket snd and rcv buffer lengths | |
362 | */ | |
0f0257ea TT |
363 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
364 | unsigned int rcv) | |
1da177e4 LT |
365 | { |
366 | #if 0 | |
367 | mm_segment_t oldfs; | |
368 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
369 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
370 | (char*)&snd, sizeof(snd)); | |
371 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
372 | (char*)&rcv, sizeof(rcv)); | |
373 | #else | |
374 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
375 | * which isn't acceptable. Until that is made conditional | |
376 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
377 | * DaveM said I could! | |
378 | */ | |
379 | lock_sock(sock->sk); | |
380 | sock->sk->sk_sndbuf = snd * 2; | |
381 | sock->sk->sk_rcvbuf = rcv * 2; | |
7f421835 | 382 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; |
47fcb03f | 383 | sock->sk->sk_write_space(sock->sk); |
1da177e4 LT |
384 | release_sock(sock->sk); |
385 | #endif | |
386 | } | |
387 | /* | |
388 | * INET callback when data has been received on the socket. | |
389 | */ | |
0f0257ea | 390 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 391 | { |
939bb7ef | 392 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 393 | |
939bb7ef NB |
394 | if (svsk) { |
395 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
396 | svsk, sk, count, |
397 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
398 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 399 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 400 | } |
1da177e4 LT |
401 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
402 | wake_up_interruptible(sk->sk_sleep); | |
403 | } | |
404 | ||
405 | /* | |
406 | * INET callback when space is newly available on the socket. | |
407 | */ | |
0f0257ea | 408 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
409 | { |
410 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
411 | ||
412 | if (svsk) { | |
413 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 414 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 415 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
416 | } |
417 | ||
418 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 419 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
420 | svsk); |
421 | wake_up_interruptible(sk->sk_sleep); | |
422 | } | |
423 | } | |
424 | ||
47fcb03f TM |
425 | static void svc_tcp_write_space(struct sock *sk) |
426 | { | |
427 | struct socket *sock = sk->sk_socket; | |
428 | ||
429 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock) | |
430 | clear_bit(SOCK_NOSPACE, &sock->flags); | |
431 | svc_write_space(sk); | |
432 | } | |
433 | ||
7702ce40 CL |
434 | /* |
435 | * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo | |
436 | */ | |
437 | static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, | |
438 | struct cmsghdr *cmh) | |
439 | { | |
440 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
441 | if (cmh->cmsg_type != IP_PKTINFO) | |
442 | return 0; | |
443 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
444 | return 1; | |
445 | } | |
446 | ||
447 | /* | |
448 | * See net/ipv6/datagram.c : datagram_recv_ctl | |
449 | */ | |
450 | static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, | |
451 | struct cmsghdr *cmh) | |
452 | { | |
453 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
454 | if (cmh->cmsg_type != IPV6_PKTINFO) | |
455 | return 0; | |
456 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
457 | return 1; | |
458 | } | |
459 | ||
9dbc240f TT |
460 | /* |
461 | * Copy the UDP datagram's destination address to the rqstp structure. | |
462 | * The 'destination' address in this case is the address to which the | |
463 | * peer sent the datagram, i.e. our local address. For multihomed | |
464 | * hosts, this can change from msg to msg. Note that only the IP | |
465 | * address changes, the port number should remain the same. | |
466 | */ | |
7702ce40 CL |
467 | static int svc_udp_get_dest_address(struct svc_rqst *rqstp, |
468 | struct cmsghdr *cmh) | |
95756482 | 469 | { |
7702ce40 CL |
470 | switch (cmh->cmsg_level) { |
471 | case SOL_IP: | |
472 | return svc_udp_get_dest_address4(rqstp, cmh); | |
473 | case SOL_IPV6: | |
474 | return svc_udp_get_dest_address6(rqstp, cmh); | |
95756482 | 475 | } |
7702ce40 CL |
476 | |
477 | return 0; | |
95756482 CL |
478 | } |
479 | ||
1da177e4 LT |
480 | /* |
481 | * Receive a datagram from a UDP socket. | |
482 | */ | |
0f0257ea | 483 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 484 | { |
57b1d3ba TT |
485 | struct svc_sock *svsk = |
486 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 487 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 488 | struct sk_buff *skb; |
bc375ea7 DM |
489 | union { |
490 | struct cmsghdr hdr; | |
491 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
492 | } buffer; | |
493 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
494 | struct msghdr msg = { |
495 | .msg_name = svc_addr(rqstp), | |
496 | .msg_control = cmh, | |
497 | .msg_controllen = sizeof(buffer), | |
498 | .msg_flags = MSG_DONTWAIT, | |
499 | }; | |
abc5c44d CL |
500 | size_t len; |
501 | int err; | |
1da177e4 | 502 | |
02fc6c36 | 503 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
504 | /* udp sockets need large rcvbuf as all pending |
505 | * requests are still in that buffer. sndbuf must | |
506 | * also be large enough that there is enough space | |
3262c816 GB |
507 | * for one reply per thread. We count all threads |
508 | * rather than threads in a particular pool, which | |
509 | * provides an upper bound on the number of threads | |
510 | * which will access the socket. | |
1da177e4 LT |
511 | */ |
512 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
513 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
514 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 515 | |
02fc6c36 | 516 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
517 | skb = NULL; |
518 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
519 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
520 | if (err >= 0) | |
521 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
522 | ||
523 | if (skb == NULL) { | |
524 | if (err != -EAGAIN) { | |
525 | /* possibly an icmp error */ | |
526 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 527 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 528 | } |
a6046f71 | 529 | svc_xprt_received(&svsk->sk_xprt); |
05ed690e | 530 | return -EAGAIN; |
1da177e4 | 531 | } |
9dbc240f | 532 | len = svc_addr_len(svc_addr(rqstp)); |
abc5c44d CL |
533 | if (len == 0) |
534 | return -EAFNOSUPPORT; | |
9dbc240f | 535 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
536 | if (skb->tstamp.tv64 == 0) { |
537 | skb->tstamp = ktime_get_real(); | |
cca5172a | 538 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
539 | need that much accuracy */ |
540 | } | |
b7aa0bf7 | 541 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 542 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 LT |
543 | |
544 | /* | |
545 | * Maybe more packets - kick another thread ASAP. | |
546 | */ | |
a6046f71 | 547 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
548 | |
549 | len = skb->len - sizeof(struct udphdr); | |
550 | rqstp->rq_arg.len = len; | |
551 | ||
95756482 | 552 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 553 | |
7702ce40 | 554 | if (!svc_udp_get_dest_address(rqstp, cmh)) { |
7a37f578 | 555 | if (net_ratelimit()) |
7702ce40 CL |
556 | printk(KERN_WARNING |
557 | "svc: received unknown control message %d/%d; " | |
558 | "dropping RPC reply datagram\n", | |
559 | cmh->cmsg_level, cmh->cmsg_type); | |
7a37f578 N |
560 | skb_free_datagram(svsk->sk_sk, skb); |
561 | return 0; | |
562 | } | |
1da177e4 LT |
563 | |
564 | if (skb_is_nonlinear(skb)) { | |
565 | /* we have to copy */ | |
566 | local_bh_disable(); | |
567 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
568 | local_bh_enable(); | |
569 | /* checksum error */ | |
570 | skb_free_datagram(svsk->sk_sk, skb); | |
571 | return 0; | |
572 | } | |
573 | local_bh_enable(); | |
cca5172a | 574 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
575 | } else { |
576 | /* we can use it in-place */ | |
0f0257ea TT |
577 | rqstp->rq_arg.head[0].iov_base = skb->data + |
578 | sizeof(struct udphdr); | |
1da177e4 | 579 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 HX |
580 | if (skb_checksum_complete(skb)) { |
581 | skb_free_datagram(svsk->sk_sk, skb); | |
582 | return 0; | |
1da177e4 | 583 | } |
5148bf4e | 584 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
585 | } |
586 | ||
587 | rqstp->rq_arg.page_base = 0; | |
588 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
589 | rqstp->rq_arg.head[0].iov_len = len; | |
590 | rqstp->rq_arg.page_len = 0; | |
44524359 | 591 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
592 | } else { |
593 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 594 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 595 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
596 | } |
597 | ||
598 | if (serv->sv_stats) | |
599 | serv->sv_stats->netudpcnt++; | |
600 | ||
601 | return len; | |
602 | } | |
603 | ||
604 | static int | |
605 | svc_udp_sendto(struct svc_rqst *rqstp) | |
606 | { | |
607 | int error; | |
608 | ||
609 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
610 | if (error == -ECONNREFUSED) | |
611 | /* ICMP error on earlier request. */ | |
612 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
613 | ||
614 | return error; | |
615 | } | |
616 | ||
e831fe65 TT |
617 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
618 | { | |
619 | } | |
620 | ||
323bee32 TT |
621 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
622 | { | |
623 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 624 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
625 | unsigned long required; |
626 | ||
627 | /* | |
628 | * Set the SOCK_NOSPACE flag before checking the available | |
629 | * sock space. | |
630 | */ | |
631 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 632 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
633 | if (required*2 > sock_wspace(svsk->sk_sk)) |
634 | return 0; | |
635 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
636 | return 1; | |
637 | } | |
638 | ||
38a417cc TT |
639 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
640 | { | |
641 | BUG(); | |
642 | return NULL; | |
643 | } | |
644 | ||
b700cbb1 TT |
645 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
646 | struct sockaddr *sa, int salen, | |
647 | int flags) | |
648 | { | |
649 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); | |
650 | } | |
651 | ||
360d8738 | 652 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 653 | .xpo_create = svc_udp_create, |
5d137990 TT |
654 | .xpo_recvfrom = svc_udp_recvfrom, |
655 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 656 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
657 | .xpo_detach = svc_sock_detach, |
658 | .xpo_free = svc_sock_free, | |
e831fe65 | 659 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 660 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 661 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
662 | }; |
663 | ||
664 | static struct svc_xprt_class svc_udp_class = { | |
665 | .xcl_name = "udp", | |
b700cbb1 | 666 | .xcl_owner = THIS_MODULE, |
360d8738 | 667 | .xcl_ops = &svc_udp_ops, |
49023155 | 668 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
669 | }; |
670 | ||
bb5cf160 | 671 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 672 | { |
7702ce40 | 673 | int err, level, optname, one = 1; |
7a37f578 | 674 | |
bb5cf160 | 675 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 676 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
677 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
678 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
679 | |
680 | /* initialise setting must have enough space to | |
cca5172a | 681 | * receive and respond to one request. |
1da177e4 LT |
682 | * svc_udp_recvfrom will re-adjust if necessary |
683 | */ | |
684 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
685 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
686 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 687 | |
0f0257ea TT |
688 | /* data might have come in before data_ready set up */ |
689 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 690 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 691 | |
7a37f578 | 692 | /* make sure we get destination address info */ |
7702ce40 CL |
693 | switch (svsk->sk_sk->sk_family) { |
694 | case AF_INET: | |
695 | level = SOL_IP; | |
696 | optname = IP_PKTINFO; | |
697 | break; | |
698 | case AF_INET6: | |
699 | level = SOL_IPV6; | |
700 | optname = IPV6_RECVPKTINFO; | |
701 | break; | |
702 | default: | |
703 | BUG(); | |
704 | } | |
705 | err = kernel_setsockopt(svsk->sk_sock, level, optname, | |
706 | (char *)&one, sizeof(one)); | |
707 | dprintk("svc: kernel_setsockopt returned %d\n", err); | |
1da177e4 LT |
708 | } |
709 | ||
710 | /* | |
711 | * A data_ready event on a listening socket means there's a connection | |
712 | * pending. Do not use state_change as a substitute for it. | |
713 | */ | |
0f0257ea | 714 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 715 | { |
939bb7ef | 716 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
717 | |
718 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 719 | sk, sk->sk_state); |
1da177e4 | 720 | |
939bb7ef NB |
721 | /* |
722 | * This callback may called twice when a new connection | |
723 | * is established as a child socket inherits everything | |
724 | * from a parent LISTEN socket. | |
725 | * 1) data_ready method of the parent socket will be called | |
726 | * when one of child sockets become ESTABLISHED. | |
727 | * 2) data_ready method of the child socket may be called | |
728 | * when it receives data before the socket is accepted. | |
729 | * In case of 2, we should ignore it silently. | |
730 | */ | |
731 | if (sk->sk_state == TCP_LISTEN) { | |
732 | if (svsk) { | |
02fc6c36 | 733 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 734 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
735 | } else |
736 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 737 | } |
939bb7ef | 738 | |
1da177e4 LT |
739 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
740 | wake_up_interruptible_all(sk->sk_sleep); | |
741 | } | |
742 | ||
743 | /* | |
744 | * A state change on a connected socket means it's dying or dead. | |
745 | */ | |
0f0257ea | 746 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 747 | { |
939bb7ef | 748 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
749 | |
750 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 751 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 752 | |
939bb7ef | 753 | if (!svsk) |
1da177e4 | 754 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 755 | else { |
02fc6c36 | 756 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 757 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 758 | } |
1da177e4 LT |
759 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
760 | wake_up_interruptible_all(sk->sk_sleep); | |
761 | } | |
762 | ||
0f0257ea | 763 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 764 | { |
939bb7ef | 765 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
766 | |
767 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
768 | sk, sk->sk_user_data); |
769 | if (svsk) { | |
02fc6c36 | 770 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 771 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 772 | } |
1da177e4 LT |
773 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
774 | wake_up_interruptible(sk->sk_sleep); | |
775 | } | |
776 | ||
777 | /* | |
778 | * Accept a TCP connection | |
779 | */ | |
38a417cc | 780 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 781 | { |
38a417cc | 782 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
783 | struct sockaddr_storage addr; |
784 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 785 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
786 | struct socket *sock = svsk->sk_sock; |
787 | struct socket *newsock; | |
1da177e4 LT |
788 | struct svc_sock *newsvsk; |
789 | int err, slen; | |
5216a8e7 | 790 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
791 | |
792 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
793 | if (!sock) | |
38a417cc | 794 | return NULL; |
1da177e4 | 795 | |
02fc6c36 | 796 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
797 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
798 | if (err < 0) { | |
1da177e4 LT |
799 | if (err == -ENOMEM) |
800 | printk(KERN_WARNING "%s: no more sockets!\n", | |
801 | serv->sv_name); | |
e6242e92 | 802 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
803 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
804 | serv->sv_name, -err); | |
38a417cc | 805 | return NULL; |
1da177e4 | 806 | } |
02fc6c36 | 807 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 808 | |
cdd88b9f | 809 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
810 | if (err < 0) { |
811 | if (net_ratelimit()) | |
812 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
813 | serv->sv_name, -err); | |
814 | goto failed; /* aborted connection or whatever */ | |
815 | } | |
816 | ||
817 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
818 | * hosts here, but when we get encryption, the IP of the host won't |
819 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 820 | */ |
cdd88b9f | 821 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 822 | dprintk(KERN_WARNING |
ad06e4bd | 823 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 824 | serv->sv_name, |
cdd88b9f | 825 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 826 | } |
ad06e4bd | 827 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 828 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
829 | |
830 | /* make sure that a write doesn't block forever when | |
831 | * low on memory | |
832 | */ | |
833 | newsock->sk->sk_sndtimeo = HZ*30; | |
834 | ||
6b174337 CL |
835 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
836 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 837 | goto failed; |
9dbc240f | 838 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
839 | err = kernel_getsockname(newsock, sin, &slen); |
840 | if (unlikely(err < 0)) { | |
841 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
842 | slen = offsetof(struct sockaddr, sa_data); | |
843 | } | |
9dbc240f | 844 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 845 | |
f9f3cc4f TT |
846 | if (serv->sv_stats) |
847 | serv->sv_stats->nettcpconn++; | |
848 | ||
849 | return &newsvsk->sk_xprt; | |
850 | ||
851 | failed: | |
852 | sock_release(newsock); | |
853 | return NULL; | |
854 | } | |
855 | ||
1da177e4 LT |
856 | /* |
857 | * Receive data from a TCP socket. | |
858 | */ | |
0f0257ea | 859 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 860 | { |
57b1d3ba TT |
861 | struct svc_sock *svsk = |
862 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 863 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 864 | int len; |
3cc03b16 | 865 | struct kvec *vec; |
1da177e4 LT |
866 | int pnum, vlen; |
867 | ||
868 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
02fc6c36 TT |
869 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), |
870 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
871 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1da177e4 | 872 | |
7f421835 BF |
873 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
874 | /* sndbuf needs to have room for one request | |
875 | * per thread, otherwise we can stall even when the | |
876 | * network isn't a bottleneck. | |
877 | * | |
878 | * We count all threads rather than threads in a | |
879 | * particular pool, which provides an upper bound | |
880 | * on the number of threads which will access the socket. | |
881 | * | |
882 | * rcvbuf just needs to be able to hold a few requests. | |
883 | * Normally they will be removed from the queue | |
884 | * as soon a a complete request arrives. | |
885 | */ | |
886 | svc_sock_setbufsize(svsk->sk_sock, | |
887 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, | |
888 | 3 * serv->sv_max_mesg); | |
889 | ||
02fc6c36 | 890 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
891 | |
892 | /* Receive data. If we haven't got the record length yet, get | |
893 | * the next four bytes. Otherwise try to gobble up as much as | |
894 | * possible up to the complete record length. | |
895 | */ | |
c0401ea0 CL |
896 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
897 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; | |
1da177e4 LT |
898 | struct kvec iov; |
899 | ||
900 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
901 | iov.iov_len = want; | |
902 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
903 | goto error; | |
904 | svsk->sk_tcplen += len; | |
905 | ||
906 | if (len < want) { | |
c0401ea0 CL |
907 | dprintk("svc: short recvfrom while reading record " |
908 | "length (%d of %d)\n", len, want); | |
a6046f71 | 909 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
910 | return -EAGAIN; /* record header not complete */ |
911 | } | |
912 | ||
913 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 914 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
915 | /* FIXME: technically, a record can be fragmented, |
916 | * and non-terminal fragments will not have the top | |
917 | * bit set in the fragment length header. | |
918 | * But apparently no known nfs clients send fragmented | |
919 | * records. */ | |
34e9a63b | 920 | if (net_ratelimit()) |
c0401ea0 CL |
921 | printk(KERN_NOTICE "RPC: multiple fragments " |
922 | "per record not supported\n"); | |
1da177e4 LT |
923 | goto err_delete; |
924 | } | |
c0401ea0 | 925 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 926 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 927 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 928 | if (net_ratelimit()) |
c0401ea0 CL |
929 | printk(KERN_NOTICE "RPC: " |
930 | "fragment too large: 0x%08lx\n", | |
931 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
932 | goto err_delete; |
933 | } | |
934 | } | |
935 | ||
936 | /* Check whether enough data is available */ | |
937 | len = svc_recv_available(svsk); | |
938 | if (len < 0) | |
939 | goto error; | |
940 | ||
941 | if (len < svsk->sk_reclen) { | |
942 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
943 | len, svsk->sk_reclen); | |
a6046f71 | 944 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
945 | return -EAGAIN; /* record not complete */ |
946 | } | |
947 | len = svsk->sk_reclen; | |
02fc6c36 | 948 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 949 | |
3cc03b16 | 950 | vec = rqstp->rq_vec; |
1da177e4 LT |
951 | vec[0] = rqstp->rq_arg.head[0]; |
952 | vlen = PAGE_SIZE; | |
953 | pnum = 1; | |
954 | while (vlen < len) { | |
44524359 | 955 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
956 | vec[pnum].iov_len = PAGE_SIZE; |
957 | pnum++; | |
958 | vlen += PAGE_SIZE; | |
959 | } | |
44524359 | 960 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
961 | |
962 | /* Now receive data */ | |
963 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
964 | if (len < 0) | |
965 | goto error; | |
966 | ||
967 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
968 | rqstp->rq_arg.len = len; | |
969 | rqstp->rq_arg.page_base = 0; | |
970 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
971 | rqstp->rq_arg.head[0].iov_len = len; | |
972 | rqstp->rq_arg.page_len = 0; | |
973 | } else { | |
974 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
975 | } | |
976 | ||
5148bf4e | 977 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
978 | rqstp->rq_prot = IPPROTO_TCP; |
979 | ||
980 | /* Reset TCP read info */ | |
981 | svsk->sk_reclen = 0; | |
982 | svsk->sk_tcplen = 0; | |
983 | ||
9dbc240f | 984 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
a6046f71 | 985 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
986 | if (serv->sv_stats) |
987 | serv->sv_stats->nettcpcnt++; | |
988 | ||
989 | return len; | |
990 | ||
991 | err_delete: | |
02fc6c36 | 992 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
993 | return -EAGAIN; |
994 | ||
995 | error: | |
996 | if (len == -EAGAIN) { | |
997 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
a6046f71 | 998 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
999 | } else { |
1000 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
bb5cf160 | 1001 | svsk->sk_xprt.xpt_server->sv_name, -len); |
93fbf1a5 | 1002 | goto err_delete; |
1da177e4 LT |
1003 | } |
1004 | ||
1005 | return len; | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * Send out data on TCP socket. | |
1010 | */ | |
0f0257ea | 1011 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
1012 | { |
1013 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1014 | int sent; | |
d8ed029d | 1015 | __be32 reclen; |
1da177e4 LT |
1016 | |
1017 | /* Set up the first element of the reply kvec. | |
1018 | * Any other kvecs that may be in use have been taken | |
1019 | * care of by the server implementation itself. | |
1020 | */ | |
1021 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1022 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1023 | ||
57b1d3ba | 1024 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
1025 | return -ENOTCONN; |
1026 | ||
1027 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1028 | if (sent != xbufp->len) { | |
0f0257ea TT |
1029 | printk(KERN_NOTICE |
1030 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
1031 | "- shutting down socket\n", | |
57b1d3ba | 1032 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
1033 | (sent<0)?"got error":"sent only", |
1034 | sent, xbufp->len); | |
57b1d3ba | 1035 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 1036 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
1037 | sent = -EAGAIN; |
1038 | } | |
1039 | return sent; | |
1040 | } | |
1041 | ||
e831fe65 TT |
1042 | /* |
1043 | * Setup response header. TCP has a 4B record length field. | |
1044 | */ | |
1045 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1046 | { | |
1047 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1048 | ||
1049 | /* tcp needs a space for the record length... */ | |
1050 | svc_putnl(resv, 0); | |
1051 | } | |
1052 | ||
323bee32 TT |
1053 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
1054 | { | |
57b1d3ba | 1055 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
47fcb03f | 1056 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 | 1057 | int required; |
323bee32 | 1058 | |
47fcb03f TM |
1059 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) |
1060 | return 1; | |
1061 | required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; | |
1062 | if (sk_stream_wspace(svsk->sk_sk) >= required) | |
1063 | return 1; | |
323bee32 | 1064 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
47fcb03f | 1065 | return 0; |
323bee32 TT |
1066 | } |
1067 | ||
b700cbb1 TT |
1068 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
1069 | struct sockaddr *sa, int salen, | |
1070 | int flags) | |
1071 | { | |
1072 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); | |
1073 | } | |
1074 | ||
360d8738 | 1075 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1076 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1077 | .xpo_recvfrom = svc_tcp_recvfrom, |
1078 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1079 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1080 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1081 | .xpo_free = svc_sock_free, |
e831fe65 | 1082 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1083 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1084 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1085 | }; |
1086 | ||
1087 | static struct svc_xprt_class svc_tcp_class = { | |
1088 | .xcl_name = "tcp", | |
b700cbb1 | 1089 | .xcl_owner = THIS_MODULE, |
360d8738 | 1090 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1091 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1092 | }; |
1093 | ||
1094 | void svc_init_xprt_sock(void) | |
1095 | { | |
1096 | svc_reg_xprt_class(&svc_tcp_class); | |
1097 | svc_reg_xprt_class(&svc_udp_class); | |
1098 | } | |
1099 | ||
1100 | void svc_cleanup_xprt_sock(void) | |
1101 | { | |
1102 | svc_unreg_xprt_class(&svc_tcp_class); | |
1103 | svc_unreg_xprt_class(&svc_udp_class); | |
1104 | } | |
1105 | ||
bb5cf160 | 1106 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1107 | { |
1108 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1109 | |
bb5cf160 | 1110 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1111 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1112 | if (sk->sk_state == TCP_LISTEN) { |
1113 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1114 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1115 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1116 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1117 | } else { |
1118 | dprintk("setting up TCP socket for reading\n"); | |
1119 | sk->sk_state_change = svc_tcp_state_change; | |
1120 | sk->sk_data_ready = svc_tcp_data_ready; | |
47fcb03f | 1121 | sk->sk_write_space = svc_tcp_write_space; |
1da177e4 LT |
1122 | |
1123 | svsk->sk_reclen = 0; | |
1124 | svsk->sk_tcplen = 0; | |
1125 | ||
b7872fe8 | 1126 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1127 | |
7f421835 BF |
1128 | /* initialise setting must have enough space to |
1129 | * receive and respond to one request. | |
1130 | * svc_tcp_recvfrom will re-adjust if necessary | |
1131 | */ | |
1132 | svc_sock_setbufsize(svsk->sk_sock, | |
1133 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, | |
1134 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1135 | ||
1136 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 1137 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1138 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1139 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1140 | } |
1141 | } | |
1142 | ||
0f0257ea | 1143 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1144 | { |
1145 | /* | |
1146 | * The number of server threads has changed. Update | |
1147 | * rcvbuf and sndbuf accordingly on all sockets | |
1148 | */ | |
1149 | struct list_head *le; | |
1150 | ||
1151 | spin_lock_bh(&serv->sv_lock); | |
1152 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1153 | struct svc_sock *svsk = |
7a182083 | 1154 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1155 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1156 | } |
1157 | list_for_each(le, &serv->sv_tempsocks) { | |
1158 | struct svc_sock *svsk = | |
7a182083 | 1159 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1160 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1161 | } |
1162 | spin_unlock_bh(&serv->sv_lock); | |
1163 | } | |
24c3767e | 1164 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1165 | |
1da177e4 LT |
1166 | /* |
1167 | * Initialize socket for RPC use and create svc_sock struct | |
1168 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1169 | */ | |
6b174337 CL |
1170 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1171 | struct socket *sock, | |
1172 | int *errp, int flags) | |
1da177e4 LT |
1173 | { |
1174 | struct svc_sock *svsk; | |
1175 | struct sock *inet; | |
6b174337 | 1176 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1177 | |
1178 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1179 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1180 | *errp = -ENOMEM; |
1181 | return NULL; | |
1182 | } | |
1da177e4 LT |
1183 | |
1184 | inet = sock->sk; | |
1185 | ||
1186 | /* Register socket with portmapper */ | |
1187 | if (*errp >= 0 && pmap_register) | |
baf01caf | 1188 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
1da177e4 LT |
1189 | ntohs(inet_sk(inet)->sport)); |
1190 | ||
1191 | if (*errp < 0) { | |
1192 | kfree(svsk); | |
1193 | return NULL; | |
1194 | } | |
1195 | ||
1da177e4 LT |
1196 | inet->sk_user_data = svsk; |
1197 | svsk->sk_sock = sock; | |
1198 | svsk->sk_sk = inet; | |
1199 | svsk->sk_ostate = inet->sk_state_change; | |
1200 | svsk->sk_odata = inet->sk_data_ready; | |
1201 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1202 | |
1203 | /* Initialize the socket */ | |
1204 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1205 | svc_udp_init(svsk, serv); |
7f421835 | 1206 | else |
bb5cf160 | 1207 | svc_tcp_init(svsk, serv); |
1da177e4 | 1208 | |
1da177e4 LT |
1209 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1210 | svsk, svsk->sk_sk); | |
1211 | ||
1da177e4 LT |
1212 | return svsk; |
1213 | } | |
1214 | ||
bfba9ab4 CL |
1215 | /** |
1216 | * svc_addsock - add a listener socket to an RPC service | |
1217 | * @serv: pointer to RPC service to which to add a new listener | |
1218 | * @fd: file descriptor of the new listener | |
1219 | * @name_return: pointer to buffer to fill in with name of listener | |
1220 | * @len: size of the buffer | |
1221 | * | |
1222 | * Fills in socket name and returns positive length of name if successful. | |
1223 | * Name is terminated with '\n'. On error, returns a negative errno | |
1224 | * value. | |
1225 | */ | |
1226 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1227 | const size_t len) | |
b41b66d6 N |
1228 | { |
1229 | int err = 0; | |
1230 | struct socket *so = sockfd_lookup(fd, &err); | |
1231 | struct svc_sock *svsk = NULL; | |
1232 | ||
1233 | if (!so) | |
1234 | return err; | |
1235 | if (so->sk->sk_family != AF_INET) | |
1236 | err = -EAFNOSUPPORT; | |
1237 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1238 | so->sk->sk_protocol != IPPROTO_UDP) | |
1239 | err = -EPROTONOSUPPORT; | |
1240 | else if (so->state > SS_UNCONNECTED) | |
1241 | err = -EISCONN; | |
1242 | else { | |
2da2c21d TT |
1243 | if (!try_module_get(THIS_MODULE)) |
1244 | err = -ENOENT; | |
1245 | else | |
1246 | svsk = svc_setup_socket(serv, so, &err, | |
1247 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1248 | if (svsk) { |
9dbc240f TT |
1249 | struct sockaddr_storage addr; |
1250 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1251 | int salen; | |
1252 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1253 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1254 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1255 | spin_lock_bh(&serv->sv_lock); | |
1256 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1257 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1258 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1259 | err = 0; |
2da2c21d TT |
1260 | } else |
1261 | module_put(THIS_MODULE); | |
b41b66d6 N |
1262 | } |
1263 | if (err) { | |
1264 | sockfd_put(so); | |
1265 | return err; | |
1266 | } | |
e7942b9f | 1267 | return svc_one_sock_name(svsk, name_return, len); |
b41b66d6 N |
1268 | } |
1269 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1270 | ||
1da177e4 LT |
1271 | /* |
1272 | * Create socket for RPC service. | |
1273 | */ | |
b700cbb1 TT |
1274 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1275 | int protocol, | |
1276 | struct sockaddr *sin, int len, | |
1277 | int flags) | |
1da177e4 LT |
1278 | { |
1279 | struct svc_sock *svsk; | |
1280 | struct socket *sock; | |
1281 | int error; | |
1282 | int type; | |
9dbc240f TT |
1283 | struct sockaddr_storage addr; |
1284 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1285 | int newlen; | |
c69da774 TM |
1286 | int family; |
1287 | int val; | |
5216a8e7 | 1288 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1289 | |
ad06e4bd CL |
1290 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1291 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1292 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1293 | |
1294 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1295 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1296 | "sockets supported\n"); | |
b700cbb1 | 1297 | return ERR_PTR(-EINVAL); |
1da177e4 | 1298 | } |
c69da774 | 1299 | |
1da177e4 | 1300 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1301 | switch (sin->sa_family) { |
1302 | case AF_INET6: | |
1303 | family = PF_INET6; | |
1304 | break; | |
1305 | case AF_INET: | |
1306 | family = PF_INET; | |
1307 | break; | |
1308 | default: | |
1309 | return ERR_PTR(-EINVAL); | |
1310 | } | |
1da177e4 | 1311 | |
c69da774 | 1312 | error = sock_create_kern(family, type, protocol, &sock); |
77f1f67a | 1313 | if (error < 0) |
b700cbb1 | 1314 | return ERR_PTR(error); |
1da177e4 | 1315 | |
ed07536e PZ |
1316 | svc_reclassify_socket(sock); |
1317 | ||
c69da774 TM |
1318 | /* |
1319 | * If this is an PF_INET6 listener, we want to avoid | |
1320 | * getting requests from IPv4 remotes. Those should | |
1321 | * be shunted to a PF_INET listener via rpcbind. | |
1322 | */ | |
1323 | val = 1; | |
1324 | if (family == PF_INET6) | |
1325 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1326 | (char *)&val, sizeof(val)); | |
1327 | ||
18114746 | 1328 | if (type == SOCK_STREAM) |
77f1f67a CL |
1329 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1330 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1331 | if (error < 0) |
1332 | goto bummer; | |
1da177e4 | 1333 | |
9dbc240f TT |
1334 | newlen = len; |
1335 | error = kernel_getsockname(sock, newsin, &newlen); | |
1336 | if (error < 0) | |
1337 | goto bummer; | |
1338 | ||
1da177e4 | 1339 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1340 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1341 | goto bummer; |
1342 | } | |
1343 | ||
e79eff1f | 1344 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1345 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1346 | return (struct svc_xprt *)svsk; |
e79eff1f | 1347 | } |
1da177e4 LT |
1348 | |
1349 | bummer: | |
1350 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1351 | sock_release(sock); | |
b700cbb1 | 1352 | return ERR_PTR(error); |
1da177e4 LT |
1353 | } |
1354 | ||
755cceab TT |
1355 | /* |
1356 | * Detach the svc_sock from the socket so that no | |
1357 | * more callbacks occur. | |
1358 | */ | |
1359 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1360 | { | |
1361 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1362 | struct sock *sk = svsk->sk_sk; | |
1363 | ||
1364 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1365 | ||
1366 | /* put back the old socket callbacks */ | |
1367 | sk->sk_state_change = svsk->sk_ostate; | |
1368 | sk->sk_data_ready = svsk->sk_odata; | |
1369 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 TM |
1370 | |
1371 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
1372 | wake_up_interruptible(sk->sk_sleep); | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Disconnect the socket, and reset the callbacks | |
1377 | */ | |
1378 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1379 | { | |
1380 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1381 | ||
1382 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1383 | ||
1384 | svc_sock_detach(xprt); | |
1385 | ||
1386 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
1387 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); | |
755cceab TT |
1388 | } |
1389 | ||
1390 | /* | |
1391 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1392 | */ | |
1393 | static void svc_sock_free(struct svc_xprt *xprt) | |
1394 | { | |
1395 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1396 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1397 | ||
755cceab TT |
1398 | if (svsk->sk_sock->file) |
1399 | sockfd_put(svsk->sk_sock); | |
1400 | else | |
1401 | sock_release(svsk->sk_sock); | |
1402 | kfree(svsk); | |
1403 | } | |
7652e5a0 BH |
1404 | |
1405 | /* | |
1406 | * Create a svc_xprt. | |
1407 | * | |
1408 | * For internal use only (e.g. nfsv4.1 backchannel). | |
1409 | * Callers should typically use the xpo_create() method. | |
1410 | */ | |
1411 | struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot) | |
1412 | { | |
1413 | struct svc_sock *svsk; | |
1414 | struct svc_xprt *xprt = NULL; | |
1415 | ||
1416 | dprintk("svc: %s\n", __func__); | |
1417 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); | |
1418 | if (!svsk) | |
1419 | goto out; | |
1420 | ||
1421 | xprt = &svsk->sk_xprt; | |
1422 | if (prot == IPPROTO_TCP) | |
1423 | svc_xprt_init(&svc_tcp_class, xprt, serv); | |
1424 | else if (prot == IPPROTO_UDP) | |
1425 | svc_xprt_init(&svc_udp_class, xprt, serv); | |
1426 | else | |
1427 | BUG(); | |
1428 | out: | |
1429 | dprintk("svc: %s return %p\n", __func__, xprt); | |
1430 | return xprt; | |
1431 | } | |
1432 | EXPORT_SYMBOL_GPL(svc_sock_create); | |
1433 | ||
1434 | /* | |
1435 | * Destroy a svc_sock. | |
1436 | */ | |
1437 | void svc_sock_destroy(struct svc_xprt *xprt) | |
1438 | { | |
1439 | if (xprt) | |
1440 | kfree(container_of(xprt, struct svc_sock, sk_xprt)); | |
1441 | } | |
1442 | EXPORT_SYMBOL_GPL(svc_sock_destroy); |